English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Existing multi-scale solutions lead to a risk of just increasing thereceptive field sizes while neglecting small receptive fields. Thus, it is achallenging problem to effectively construct adaptive neural networks forrecognizing various spatial-scale objects. To tackle this issue, we firstintroduce a new attention dimension, i.e., depth, in addition to existingattention dimensions such as channel, spatial, and branch, and present a novelselective depth attention network to symmetrically handle multi-scale objectsin various vision tasks. Specifically, the blocks within each stage of a givenneural network, i.e., ResNet, output hierarchical feature maps sharing the sameresolution but with different receptive field sizes. Based on this structuralproperty, we design a stage-wise building module, namely SDA, which includes atrunk branch and a SE-like attention branch. The block outputs of the trunkbranch are fused to globally guide their depth attention allocation through theattention branch. According to the proposed attention mechanism, we candynamically select different depth features, which contributes to adaptivelyadjusting the receptive field sizes for the variable-sized input objects. Inthis way, the cross-block information interaction leads to a long-rangedependency along the depth direction. Compared with other multi-scaleapproaches, our SDA method combines multiple receptive fields from previousblocks into the stage output, thus offering a wider and richer range ofeffective receptive fields. Moreover, our method can be served as a pluggablemodule to other multi-scale networks as well as attention networks, coined asSDA-$x$Net. Their combination further extends the range of the effectivereceptive fields towards small receptive fields, enabling interpretable neuralnetworks. Our source code is available at\url{https://github.com/QingbeiGuo/SDA-xNet.git}.

SDA-$x$Net: Selective Depth Attention Networks for Adaptive Multi-scale Feature Representation